Brand Intelligence

What this is for, in one sentence: Brand Intelligence audits what AI platforms actually claim about your brand — every claim extracted, fact-checked against sources, and turned into a prioritized fix plan — so the story AI tells your buyers is one you’ve verified.

When to come here:

• Early in your setup, before you invest in new content — fix what AI already believes first
• After any major change to pricing, positioning, integrations, or company facts
• On a regular cadence, because AI answers drift as models and sources update

Priya Nair’s job is MenuPilot’s story, and the most influential re-teller of that story answers thousands of buyer questions a day without ever calling her for comment. This is the screen where she reads what it’s been saying — and finds out that some of it is confidently, sourced-ly wrong.

---

Running an analysis

The first run starts from a single button. The screen tells you what you’re committing to before you click: it audits what AI platforms claim about your brand, fact-checks every claim, and returns a prioritized fix list — takes about 5–7 minutes and uses 1 analysis from your quota. It works from your completed AI responses across all five platforms, so the richer your tracked prompt set, the more claims it has to audit.

While it runs, the progress view shows the actual pipeline: collecting your AI responses, extracting the distinct claims made about your brand, fact-checking each claim against sources, then scoring and building the action plan. Watching “31 of 48 claims checked” tick by is also a useful mental model for what this module is — not a vibe check, but a claim-by-claim audit with receipts.

Reading the results

Three numbers at the top. Claim Accuracy is the headline — the share of extracted claims that check out (here, 39 of 48 = 81.3%). Perception Alignment is a product-computed composite on a 0–100 scale, summarizing how closely AI’s overall picture matches your verified reality. Claims Flagged counts what needs your attention, split into inaccurate and unverifiable.

The Gaps tab is where the work is. Each flagged claim shows: the claim as AI states it, which engines said it, a severity level, and — the part that makes this actionable — the fact-check itself, with a confidence label and the sourced truth. “MenuPilot was founded in 2019” sits next to Founded 2021 · menupilot.io/about, high confidence, sourced. You’re never asked to take the product’s word for it.

Every row offers three responses, and the wording is deliberate:

• Confirm this is wrong — you’re confirming the claim is false. Confirmed gaps feed the action plan.
• This is actually correct — AI was right and your records were the problem; the flag clears.
• Add to fix list — route it straight into your working backlog.

Unverifiable claims — “used by over 2,000 restaurants,” a stat with no public source behind it — are labelled low confidence and wait for your confirmation rather than being auto-judged. Nobody but you knows whether that number is true.

Confirm before you act — the ten minutes that keep the plan honest. The action plan is generated only from gaps you’ve confirmed. Triage every row: confirm the genuinely wrong, clear the actually-correct, and give the unverifiables a real answer. A fix list built on unreviewed flags is just noise with a to-do format.

Unverifiable doesn’t mean false. It means no public source exists to check the claim against — which is itself a finding. If a claim about you is true and important, the fix is often to publish the source (a stat on your site, a page that states the fact) so both AI and this audit have something to check against.

The action plan

Confirmed gaps become a short, prioritized plan — here, 5 confirmed gaps consolidated into 4 actions, each with severity, a plain-language instruction stating the verified truth (down to the foundingDate: 2021 schema property), and a count of the claims it resolves. Notice the consolidation: two pricing-related gaps became one “publish an unambiguous pricing page” action, because that’s how the work actually ships.

Unverifiable claims sit in their own Needs verification bucket, explicitly waiting for your review in the Gaps tab — they don’t dilute the plan, and they don’t get forgotten.

Execution lives one module over: the Action Center generates the schema, the corrected pages, and the llms.txt that state your verified facts where machines read them.

Fixes take time to reach AI answers. Correcting your site changes the sources; the answers follow as AI systems re-crawl and update — the lag varies by engine and can run weeks. Re-run the analysis after shipping fixes to track the accuracy score, and judge the trend across runs rather than expecting next-day corrections.

Common questions

How often should I re-run it? After any batch of shipped fixes, and on a regular cadence otherwise — monthly is a sensible default for most brands. Each run takes ~5–7 minutes and uses 1 analysis from your quota, so the constraint is rarely the tool.

What’s the difference between Claim Accuracy and Perception Alignment? Accuracy is arithmetic — accurate claims over total claims. Alignment is a broader composite index (0–100) of how well AI’s picture matches reality overall. Watch accuracy for the direct effect of your fact fixes; watch alignment for the bigger drift.

A claim is flagged wrong but it’s actually true — what do I do? Click “This is actually correct.” The flag clears, and your accuracy picture improves without any site changes. This is also worth doing promptly — it keeps the action plan from carrying work you don’t need.

AI says something wrong about us that isn’t in the list — why? Claims are extracted from your tracked responses. If a wrong claim circulates on a question you don’t track, add that prompt (Setting Up Prompt Tracking) and re-run — the audit can only check what your prompt set surfaces.

Do the engines matter individually? Yes — each gap shows which engines make the claim, and that’s targeting information. A wrong fact confined to one engine usually traces to a source that engine favors; the fix priority and the correction route can differ accordingly.

What to read next

• Action Center — where confirmed gaps become shipped schema, pages, and llms.txt.
• Visibility Gaps — the broader backlog this module’s fact fixes slot into.
• Brand Sentiment — the tone layer: after AI’s facts are right, this is what it feels like AI says about you.